For several years now, EDF has been implementing an ambitious strategy aimed at integrating artificial intelligence into the core of its engineering and operations. The applications are varied, ranging from “pure AI” use cases—such as image analysis for inspection, time-series processing for predictive maintenance, or multimodal problem-solving—to the acceleration of physical simulations via surrogate modeling. This latter approach is a key enabler for digital twins and design support. The challenge common to all these projects lies in computational constraints, which require rigorous choices regarding AI architecture and hardware infrastructure for large-scale deployment. In response, EDF’s R&D focuses on optimizing hardware infrastructure (machine management, GPUs) and developing frugal, robust, and reliable AI models tailored to the critical constraints of the energy sector.
This presentation will offer an overview of this strategy, its successes, and its challenges. |